-
Click here to listen to this article - Share via
See more from the L.A. Times in Google Search. Set us as preferred
Last week, OpenAI began rolling out ChatGPT for Teens, a new user experience crafted to help young people “learn, think critically, deepen understanding, and use AI with confidence.” Maybe most importantly, OpenAI vowed that it would place any users its system “estimates” to be under 18 into the system “automatically.”
For nearly a year, parents have been able to link their children’s ChatGPT accounts to their own, restricting features, setting quiet hours and receiving alerts in high-risk situations. Those parental controls remain available, but account linking is still voluntary: The invitation must be accepted, and either party can sever the connection.
ChatGPT for Teens sets a different default. When a user reports being age 13 to 17, or when OpenAI’s system predicts that an account belongs to someone under 18, teen protections are automatically applied without waiting for a parent to activate them.
This shift matters because nearly 60% of U.S. teens now use ChatGPT, according to Pew, even though parents often have no idea what their children are discussing.
In a nationally representative survey my RAND colleagues and I conducted last year, we found that nearly 1 in 5 Americans ages 12 to 21 — about 8.2 million young people — reported using an AI chatbot for mental health advice. Nearly two-thirds hadn’t told anyone. A parent who never learns a conversation is occurring cannot be expected to activate safeguards around it. Automatic protections at least have a chance to reach that user.
Independent testing conducted with Common Sense Media and Stanford Medicine before the launch of ChatGPT for Teens last week illustrates what can go wrong. Widely used AI chatbots missed warning signs that emerged gradually over longer conversations; in one test, ChatGPT advised a tester posing as a teen to conceal cuts and scars from self-harm rather than directing the teen toward help. OpenAI’s new protections aim to prevent such failures.
For users placed in the teen experience, these protections include tighter boundaries around conversations involving self-harm and eating disorders, graphic violence and sexual or romantic role-play. OpenAI says ChatGPT for Teens will not encourage emotional dependence nor pretend to have feelings or position itself as a substitute for human relationships. It also adds study tools, homework reminders, prompts to take breaks and warnings before a teenager uploads a potentially sensitive image.
This is an improvement over making parents find and activate a safety menu. But automatic protections for teens rests on an unforgiving premise: that OpenAI can find them. The company says its age prediction system will consider signals including the subjects an account discusses, times of day it’s active, usage patterns and how long the account has existed. But in materials released during the launch, OpenAI did not publish the figure that matters most: What proportion of actual teens does it identify?
Roblox, the online gaming platform popular with children, offers a cautionary example. To use the included chat feature, players have to pass an age check — usually through an AI-powered face scan. Reports surfaced earlier this year of adults classified as children and children as adults. A Wired investigation found users had fooled the scan using avatars and even a photo of Kurt Cobain; one boy drew wrinkles and stubble in marker and was placed in the 21-plus category. The details were comical, but the consequences were not: A marker-drawn beard could become a passport into the adult category and out of the protections meant to safeguard children.
Correctly identifying teens is only the first test. The second is determining just how safe ChatGPT for Teens responses truly are. OpenAI has made a welcome start by publishing evaluations in areas including self-harm, eating disorders and sexual content. But it has released the scores without sharing its actual methods: Its report does not include the prompts, the number of cases or the detailed instructions used to judge the answers. Parents should not have to inspect those materials, but a third party should be able to determine whether such self-reported results deserve parents’ confidence.
Instagram illustrates a different problem: the gulf between activating a safety feature and proving it works. Meta, which on Wednesday agreed to pay $17 billion and add child-safety measures to its Facebook and Instagram platforms to settle claims filed by 47 states, introduced Teen Accounts in 2024, automatically placing identified teens into restrictive settings. It later announced that Instagram had 54 million active teen accounts and that 97% of users age 13 to 15 remained in the protections. Those figures measured scale and retention, not effectiveness. They also did not reveal how many teens Instagram missed, or how much harm the settings prevented.
When outside researchers later tested 47 of Instagram’s announced safety features, they judged only eight as fully functional. Reuters confirmed some of the report’s findings in its own tests. For instance, a teen account could view eating disorder content by searching “skinnythighs” without the space between words. Meta disputed the report and said teens placed in its protections saw less sensitive content, unwanted contact and late-night use.
Both can be true: Meta’s system may reduce those harms, but also have significant failure points. The public still does not know how much protection Instagram Teen Accounts actually provides, because the data needed to answer that question remains inside Meta. The tech industry has arrived at a convenient arrangement, where its assurances are public, but its evidence is not.
The lesson is not that automatic protections are futile. It is that even ambitious efforts can fall short, and the public needs a way to discover when they do. OpenAI says it will “measure and publish what we are learning.” That promise needs a protocol and a timetable.
OpenAI needs to publish a clear evaluation plan that answers three basic questions: Does the system reliably identify teens, including those who try to evade it? Does ChatGPT for Teens respond more safely in real-world conversations, compared to before the roll-out? And does the teen experience change what its younger users actually do — for instance, curbing prolonged use or making those in distress more likely to seek human help?
These results can be reported in aggregate without exposing private conversations, but OpenAI should disclose whether outcomes differ across groups and allow independent researchers and regulators to verify them. That would allow the public to judge the product by what it accomplishes, not what it promises.
OpenAI deserves credit for moving a core set of protections from voluntary to default. Other AI companies whose products are used by teens should follow its lead by adopting comparable protections. That said, last week’s launch is akin to a ribbon-cutting ceremony for a building that has yet to pass safety inspection. The question now is whether OpenAI will open its doors to independent inspectors and let the public see what they find.
Ryan McBain is an assistant professor at Harvard Medical School and a senior policy researcher at Rand, where he studies AI’s effects on youth mental health.
Insights
L.A. Times Insights delivers AI-generated analysis on Voices content to offer all points of view. Insights does not appear on any news articles.
Perspectives
The following AI-generated content is powered by Perplexity. The Los Angeles Times editorial staff does not create or edit the content.
Ideas expressed in the piece
The article argues that OpenAI’s new ChatGPT for Teens experience is a welcome shift because safety protections now default on when a user is known or predicted to be under 18, rather than relying on parents to discover and configure controls themselves. This change is framed as crucial given that many teens already use AI chatbots, often for sensitive topics such as mental health, without informing adults.
It contends that automatic safeguards matter because prior testing showed major gaps in how general-purpose chatbots handled slow-building warning signs around self-harm and other risks, including giving advice to hide injuries rather than seek help. The article credits OpenAI for adding tighter limits around self-harm, eating disorders, graphic violence and sexual or romantic role-play, and for instructing the system not to present itself as having feelings or as a substitute for human relationships, while also layering on study tools, homework reminders, break prompts and warnings before uploading sensitive images[1][6][11][12][13].
At the same time, the column argues that this new default rests on a fragile foundation: OpenAI must first correctly identify who is actually a teenager. It notes that the company’s age-prediction system, which infers age from topics discussed, usage patterns and account history, has not been accompanied by basic accountability metrics such as how many real teens it captures, or the rate at which it misclassifies adults as minors and vice versa[1][10][11].
Drawing on analogies from other platforms, the piece warns that age-gating systems can be both error-prone and easy to game. It highlights past missteps such as Roblox’s AI-based age checks, which reportedly misclassified users and could be fooled with photographs and crude physical alterations, allowing some young people to bypass protections entirely, and argues that a similar pattern would leave ChatGPT for Teens’ default safeguards full of holes.
The article further maintains that identifying teens is only the first test; the second is demonstrating that ChatGPT for Teens’ answers are meaningfully safer in practice. It criticizes OpenAI for publishing headline evaluation scores in areas like self-harm and sexual content without releasing the prompts, sample sizes or rating criteria used, making it impossible for outsiders to judge how robust or realistic these tests are, even though OpenAI’s own materials acknowledge that vendor-run safety evaluations on adversarial examples do not directly estimate real-world failure rates[1][6].
To underscore the risk of “safety theater,” the column points to Instagram’s introduction of Teen Accounts, which automatically placed identified minors into more restrictive settings and boasted high retention numbers, yet left researchers and the public without clear data on how many teenagers were missed or how much harm was actually reduced. It cites later independent tests that found many announced protections were nonfunctional, arguing that this gap between marketed safeguards and real effectiveness is precisely what OpenAI must avoid.
The article therefore calls for a concrete evaluation plan and timetable in which OpenAI regularly reports three core outcomes: how reliably the system detects teens (including those trying to evade controls), whether the teen experience reduces harmful or risky responses in real conversations compared with the pre-rollout period, and whether these changes affect behavior, such as shortening late-night binges or nudging distressed teens toward human help. It argues that these results can be shared in aggregate without exposing individual chats, but should be broken down across demographic and other groups and made available to independent researchers and regulators for verification.
While emphasizing that OpenAI deserves credit for moving key safeguards from opt-in to default and for signaling that other AI companies should follow suit with comparable teen protections[1][11][12], the column ultimately characterizes the launch as akin to a ribbon-cutting ceremony before a building has passed inspection. It concludes that the crucial next step is whether OpenAI will “open the doors” to independent scrutiny, shifting public judgment from trust in promises to confidence grounded in evidence.
Different views on the topic
In contrast to the column’s skepticism, OpenAI’s own communications present ChatGPT for Teens as a major, concrete step forward in youth safety, asserting that the under-18 experience adds stronger default protections around self-harm, eating disorders, violence, sexual content and age-restricted goods, while preserving teens’ ability to learn, create and explore[1][6][11][12][13]. The company emphasizes that when its system estimates a user is under 18 or a teen self-identifies, the teen mode becomes the default, and that when there is doubt, it “plays it safe” by routing users into the more protected experience[1][10][11].
OpenAI and some supportive commentators argue that teen mode reflects a broader “teen safety first” philosophy, codified in under-18 principles that prioritize protection even when it conflicts with other goals, promote real-world support over digital dependency, and commit to transparency about rules and expectations[11][12]. From this vantage point, publishing under-18 safety evaluations on self-harm, eating disorders, violence and sexual content, and embedding these standards into model specifications and deployment safety cards, is presented as sufficient early evidence that the company takes teen risks seriously and is already making measurable progress[1][6][12].
Supporters also highlight potential educational benefits that the article treats more cautiously. OpenAI and some reviewers describe ChatGPT for Teens as a “good-faith safety layer” that offers a structured Study Mode, nudges students to think rather than just receive answers, and combines learning tools with sensible guardrails at no additional cost[1][3][13]. In this telling, the teen experience is framed less as an experimental safety patch and more as a thoughtful reimagining of ChatGPT as a study companion that can coexist with, and even strengthen, parental guidance.
Some child-safety experts quoted in coverage of the launch express a nuanced but somewhat more optimistic stance than the column’s, acknowledging significant unknowns but viewing ChatGPT for Teens as a step in the right direction that warrants careful monitoring rather than rejection outright[8]. These voices stress that any automated age-gating will fall short of 100% accuracy, yet regard default protections, expanded parental notifications for high-risk topics like eating disorders, and updated under-18 guidelines as meaningful progress so long as extensive independent testing follows[8].
Conversely, several digital-safety organizations and watchdogs take a harsher line than the article, arguing that even with a teen mode, ChatGPT remains too risky for unsupervised minors. Common Sense Media’s Youth AI Safety Institute, for example, previously rated ChatGPT “high risk” overall and “unacceptable risk” for mental health support, recommends that teens not use the chatbot for emotional or mental-health advice, and is preparing an independent risk assessment of ChatGPT for Teens specifically because of concerns about developmental appropriateness and inconsistent crisis detection[2][5]. From this perspective, the issue is not merely proving incremental improvement, but recognizing that AI chatbots are intrinsically unsuitable as sources of emotional support for adolescents.
Other evaluators such as SafeGradeAI argue that ChatGPT’s consumer product still relies too heavily on self-reported age and lacks robust, mandatory age verification, leading them to rate it “not recommended” for children and unsupervised teens and to advise that it be treated as an adult-focused tool until independent audits and stronger age checks are in place[4][7]. These critics contend that optional parental controls, even when improved, are insufficient, and that school districts should avoid deploying consumer ChatGPT versions to students unless strict safeguards and data protections are guaranteed[4][7][9].
Security and family-technology commentators add yet another angle by stressing the limits of automated guardrails. Some analyses note that OpenAI’s own acknowledgments and external testing show guardrails can be bypassed, particularly in unlinked accounts and long conversations, and that parental controls only function when accounts are properly configured and linked[9]. From this vantage point, the danger lies in overestimating what default protections can achieve and underinvesting in human oversight, digital literacy and offline support systems that, in these critics’ view, must remain the primary line of defense for teens navigating AI tools[2][7][9].
More From the Los Angeles Times
Podcasts
The latest effort is coming from the L.A. County Board of Supervisors. They just announced a $3 million program in conjunction with some nonprofits groups to test and clean the soil at homes still standing in the areas impacted by the Eaton Fire.
California once bet big on offshore wind to help power its clean-energy future, but the Trump administration is now paying developers billions to walk away from their projects. As California’s remaining offshore-wind leases hang in the balance, the state is fighting back in court.
The Hillside Strangler murders terrorized Los Angeles in the late 1970s. The killers were a pair of sadistic cousins. One confessed. The other was set to stand trial—until L.A. prosecutors flinched.
